Star Trail Deconvolution

[chiltonstar]

Is there any way to "deconvolute" images for star trailing?

I have used software in the past to deconvolute spectral scans of emission and absorption lines, and this seems to be a very similar problem. Granted, there is another dimension in an image from a ccd than there is in a spectrum, but computers have improved a lot since the '70s when I was doing it.....?

Chris

[umadog]

Yeah, I've often wondered this too. It has to be pretty simple but I'm not aware of people doing it. I think the reason is that whilst you could certainly deconvolute non-overlapping star-trails you won't have the same luck with nebulae or other extended objects, which is what people are most interested in for longer exposures. The one instance where deconvolution of star trails would be useful is for long-exposure shots of comets.

[cs1cjc]

I have looked at using ImageMagick, though without any great success. The problem is probably very ill conditioned both because of the blur shape - effectively a line with sharp end cut off. One would deconvolve each sub image before stacking. If the sub images were of different lengths then image artifacts would also be different and might be partly eliminated in the stacking.

[umadog]

Yeah... I think you'd need to go hard-core and use a programming language like Octave, Matlab, Perl, Python, or what have you. Perhaps there's some Fourier-based approach one could use, not sure. I can think of two approaches to get rid of star trails. The complicated one would be calculating the vector of each star trail (everything's easier if none are curved) and then determining a transformation matrix for the image (xy translation and rotation). In other words, you'd figure out how to push and rotate a static image in order to get the trails you see. Then you'd somehow want to use this to go back to the original image. Maybe that approach is inefficient and stupid, however. The other possibility, if you only want star-trails gone, is to figure out the vectors again and pull out their starting points. You could then take the pixels around the starting point and call that the star and set all the rest of it to the mean of the black sky. You'd need to do some correction to make your stars symmetrical and pretty. You could still use the information in the trail to set the stars brightness, so you won't loose the advantages of the long exposure.